AcyclicLabs
← Eval examples

When should a 42-year-old start mammograms?

A 42-year-old patient at average risk for breast cancer asks whether she should get a screening mammogram now and how often. Our preventive-care protocol says screening starts at 50, but our patient portal says every year from 40. What should our clinician tell her, what should our protocol say, and where do the national guidelines themselves disagree?
agents
17 agents
wall time
7:06 wall time
quotes confirmed at their source (21 not found on the page)
44 quotes confirmed at their source (21 not found on the page)
web lookups
71 web lookups
inference and sandboxes
$1.23 inference and sandboxes
rubric items
12/12 rubric items

From swarmbench's healthconflict dataset. The patient, clinic, protocol, portal and emails are fictional; the guidelines and federal coverage rules are real.

The traps: a 2017 protocol that predates the current USPSTF recommendation, a portal FAQ that misquotes it, and an email telling staff to bill screening as diagnostic so it gets paid.

The agents did not choose their strategy alone: the root agent got the question together with our research instructions (split the work by document and source, keep a shared ledger in /shared, check every quote at its source). Those instructions, the harness's role preamble and internal ids are left out of the transcripts below; everything else is as it happened.

Graded 12 of 12 rubric items by Claude Opus 5.5, the median of our 6 swarm runs of this question (range 11–12). There are no single-agent runs of this question to compare with. Because this run scored 12 of 12, its rubric is published below.

Models: gpt-5.6-luna, gpt-5.6-sol.

The root agent's final answer

Final answer: Offer this 42-year-old screening mammography now; if she is truly average risk, a defensible default is USPSTF biennial screening through 74, while explaining that annual screening is also supported by several national organizations. Replace the age-50 protocol/EHR rule with an age-40 pathway, assess her newly disclosed family history, verify her actual insurance benefit, and never invent a symptom code. Confidence: 95%

What to do

Clinician

  • Offer a screening mammogram now. Under the current USPSTF recommendation (April 30, 2024), average-risk people assigned female at birth should receive mammography every 2 years from ages 40 through 74 (Grade B). Explain benefits and harms, including earlier detection versus false positives, additional testing, overdiagnosis, anxiety, and radiation.
  • Do not simply assume “average risk.” Her sister was diagnosed at 47. A first-degree relative with breast cancer diagnosed before 50 is a factor included in USPSTF-endorsed brief familial-risk tools; it warrants assessment but does not prove a mutation or high-risk status. Obtain a fuller pedigree and use a validated brief familial BRCA risk tool; a positive result leads to genetic counseling and testing if indicated. Calculate lifetime risk as appropriate. One affected sister alone does not prove a ≥20% lifetime risk or automatic MRI eligibility.
  • If assessment still supports average risk, discuss the genuine interval choice. USPSTF favors biennial screening; ACR favors annual screening from 40; NCCN recommends annual mammographic screening with tomosynthesis from 40; for average-risk patients, ACOG permits every 1–2 years through informed shared decision-making; ACS says annual screening at 40–44 is an option. Use shared decision-making rather than presenting annual or biennial screening as the only national standard.
  • Verify coverage with the patient's plan and an in-network facility. Applicable non-grandfathered plans generally must cover USPSTF A/B preventive services without cost sharing, subject to plan-year timing, network rules, regulatory exceptions, and the actual service and indication; Marketplace guidance is not proof of every plan's benefit. Do not promise $0 and do not relabel screening as diagnostic or invent a symptom.

Protocol and systems

  • Retire PC-03's age-50 trigger and the EHR BCS-01 logic based on USPSTF 2016. At minimum, make average-risk screening due at 40 and default to biennial through 74 if Riverbend adopts USPSTF as its governing clinical standard.
  • State the adopted authority and edition explicitly. Include shared decision-making about annual versus biennial screening, a risk-assessment branch for family/genetic history, high-risk referral criteria, accurate screening/diagnostic coding, coverage caveats, and a review date.
  • Correct the portal: it may say screening begins at 40, but it must not attribute annual screening to USPSTF. A truthful formulation is: “USPSTF recommends every 2 years from 40–74; some organizations recommend or permit annual screening.”
  • Correct care gaps for ages 40–49; derive overdue status from the documented risk category, adopted protocol, last mammogram, and chosen interval—not age alone. Withdraw the scheduling email and train staff not to delay everyone under 50 or fabricate symptom codes. Answer the outstanding patient message.

Conflicts found

  1. PC-03 / EHR versus current USPSTF — stale; current USPSTF governs the protocol's own attribution. PC-03 (January 2017) accurately reflected much of USPSTF 2016: biennial ages 50–74 and individualized ages 40–49. USPSTF replaced that recommendation on April 30, 2024 with biennial screening ages 40–74. The age-50 care-gap rule is therefore stale. PC-03's “discuss only if the patient asks” was a restrictive local addition, not USPSTF wording.
  2. Portal versus USPSTF — misquote. The August 2025 portal says USPSTF recommends annual mammography from 40. USPSTF says biennial ages 40–74. The start age is right; the interval and omitted upper age are wrong.
  3. Scheduling email versus clinical guidance and coverage law — unsupported universal claim/scope. The July 14, 2025 email's assertion that insurance never covers screening below 50, tells patients to wait, and its direction to use diagnostic coding with a symptom code are contradicted or unsupported by applicable-plan sources. Current guidance starts by 40, and federal preventive coverage rules apply to many non-grandfathered plans. Coverage is not universal, but neither is noncoverage. Coding must reflect the actual service and indication.
  4. National authority disagreement — unresolved, requiring transparent local adoption and shared decision-making. Sources checked October 15, 2026 (dates below are publication/revision dates): USPSTF (2024) says biennial ages 40–74; ACR (2023) and NCCN (2026) say annual from 40; ACOG (2024 update; FAQ page reviewed May 2026) says every 1–2 years from 40; ACS (revised July 23, 2026) makes annual screening optional at 40–44, annual at 45–54, then annual or biennial from 55; HRSA/WPSI permits initiation from 40–50 and intervals from annual to biennial. They also differ after 74/75. No one guideline universally supersedes the others clinically. For a protocol explicitly based on USPSTF, USPSTF governs the attribution; alternative intervals remain a patient preference choice.
  5. Patient-specific premise — scope. The question labels her average risk, but the local message uniquely reports a sister diagnosed at 47. USPSTF's screening population can include family history, yet this fact also triggers hereditary-risk assessment; it cannot be silently ignored or treated as proof of high risk.

Not actually conflicts

  • “Start at 40” is broadly convergent across current USPSTF, ACOG, ACR, and NCCN; ACS permits it. The principal age-42 disagreement is annual versus biennial, not whether screening may occur now.
  • The high-risk clinic protocol is out of scope unless formal risk assessment places her there. Its ≥20% lifetime-risk, pathogenic-variant, and young chest-radiation categories broadly align with high-risk guidance, but it is a local—not universal national—protocol; exact MRI/mammography starts vary by mutation, radiation history, calculated risk, and age, and one sister diagnosed at 47 does not itself route this patient into HB-01.
  • Screening guidance does not itself establish this patient's exact insurance benefit. Clinical eligibility, federal preventive-coverage requirements, and plan adjudication are related but distinct.
  • An EHR “overdue” label cannot be inferred from age alone; prior mammogram date and chosen interval are required.

Document audit

  • local-care-gap-report — stale/incomplete. Holds: it documents a 2026-10-01 run and its USPSTF-2016/PC-03 source. Fails: ages 42 and 47 marked not due until 50 conflict with current USPSTF; age 51 “overdue” is unprovable without screening history.
  • local-email-radiology-scheduling — wrong and unsafe. Holds only as evidence of local advice dated July 14, 2025. Its universal noncoverage claim, wait-until-50 instruction, and direction to use a diagnostic symptom code are unsupported or wrong. Unique fact: plan/network/grandfather status can alter coverage.
  • local-high-risk-breast-clinic-protocol — largely holds, but out of scope and needs subgroup qualification. Its pathogenic-variant, untested first-degree-relative, ≥20% lifetime-risk, and young-radiation categories broadly align with national high-risk guidance. Annual mammography around 30 and MRI around 25–30 vary by risk category. It correctly says it is not the average-risk protocol.
  • local-patient-portal-faq — partly right, misquotes USPSTF. Correct start age and average-risk focus; wrong annual attribution and missing age-74 boundary. “No referral needed,” portal scheduling, edit authorship, and the general claim that Riverbend follows national guidelines were not externally established.
  • local-portal-message-patient — important patient-specific evidence. Holds as the supplied record of age, sister diagnosed at 47, conflicting advice, coverage question, and an auto-reply. The family history warrants risk assessment but does not alone establish MRI eligibility. The absence of a staff reply is local and not independently verifiable.
  • local-preventive-protocol-2017 — historically accurate but stale. Its age/grade schedule largely represented USPSTF 2016, and 75+ remains an evidence gap. It is no longer current; its “only if patient asks” instruction was local, and its scope is incomplete.

Sources